[j-nsp] Summarize Global Table

Chris Morrow morrowc at ops-netman.net
Wed Oct 26 10:45:31 EDT 2011



On 10/26/2011 10:24 AM, Mark Tinka wrote:
> On Wednesday, October 26, 2011 09:26:11 PM Chris Morrow 
> wrote:
> 
>> As I've said in the grow-wg sessions several times (and
>> at nanog and other places) VA, and other solutions like
>> it, may be fine in some deployments, they may even save
>> you some cycle time on RP/RE/linecards, each operator
>> that uses these solutions needs to decide for themselves
>> what level of loss in granularity is acceptable and on
>> which platforms in their network they will still need to
>> carry full routes in rib + fib.
> 
> For networks that may be feeling pressure in the core (and 
> if the core is especially large that the problem is 
> magnified), this really is where MPLS can have real 
> application, as opposed to just being a buzz word for folk 
> that think it's cool.

this is a LOT of overhead though, where 'how about we just supress these
/24s k?' could work 'as well'.

> If removing BGP from your core will save you tons of $$ in 
> core router upgrades, then MPLS is certainly worth 
> considering. Of course, this assumes the network is 

again, overhead/tradeoffs... it's not a 'bad thing' just something to
keep in mind, it's not as blithe as 'sure, turn on that em-pee-ell-ess'
as I've heard many times :)

> currently already running MPLS, or does not find deploying 
> to be rocket science. If neither of those pre-conditions is 
> the case, then it's not such a hot idea, of course.
> 
> Also, it assumes your edge (particularly the Aggregation 
> nodes) can still handle a full table, which may also not be 
> the case for some networks.

yup... good thing people removed those 7500's from the edge! and
12008's! (and m40's)... oh, wait :(


More information about the juniper-nsp mailing list